Paraguay Wildfire Risk Explorer

English
Spanish French Portuguese English

Data and Methods

Technical documentation for the Paraguay Wildfire Risk Explorer

Overview

This web application provides near real-time wildfire risk analysis for the Paraguayan Chaco region. The system combines live fire detection data from NASA satellites with spatial datasets representing vulnerable populations and protected areas to assess immediate fire threats.

Data Sources

Dataset Source Type Update Frequency
Active Fire Hotspots NASA FIRMS (VIIRS S-NPP) Point Near Real-Time (3-5 hours)
Administrative Boundaries Paraguayan Statistical Institute (DGEEC) Polygon Static (2020)
Protected Areas World Database on Protected Areas (WDPA) Polygon Annual (January 2026)
Indigenous Communities Paraguayan Statistical Institute Polygon Static
Human Settlements Paraguayan Statistical Institute Point Static

Data Processing Workflow

1. Spatial Data Preparation

All spatial datasets undergo standardized preprocessing:

  • Coordinate System Standardization: All layers transformed to UTM Zone 20S (EPSG:32720)
  • Geographic Clipping: Data clipped to Paraguayan Chaco administrative boundaries (Alto Paraguay, Boquerón, Presidente Hayes)
  • Geometry Validation: Invalid geometries corrected; polygons converted to MultiPolygon format for database consistency
  • Attribute Cleaning: Unnecessary fields removed, column names standardized to lowercase

2. Database Architecture

Processed data is stored in a Supabase PostgreSQL database with the PostGIS extension:

  • Spatial indexing (GIST) on all geometry columns for optimized spatial queries
  • Custom PostgreSQL functions for region-based data filtering
  • Real-time Edge Functions for NASA FIRMS API integration

3. Live Fire Data Integration

Active fire hotspots are retrieved on-demand via NASA FIRMS API:

  • Data source: VIIRS S-NPP Near Real-Time (NRT) instrument
  • Bounding box dynamically calculated from selected Chaco region
  • CSV data converted to GeoJSON format server-side
  • Fire confidence levels (high, nominal, low) preserved for risk weighting

Analysis Methodology

Proximity-Based Risk Assessment

The application employs client-side spatial analysis using Turf.js to identify features at risk:

Distance Calculation

For each active fire point f and each target feature t (household, community, or protected area):

  1. If t is a polygon, compute its centroid as the evaluation point
  2. Calculate Euclidean distance d(f, t) in kilometers using Turf.js
  3. Classify risk level based on user-defined thresholds:
  • ■ High Risk: d ≤ High Threshold (default: 1000m)
  • ■ Moderate Risk: High Threshold < d ≤ Moderate Threshold (default: 2500m)

Deduplication

Features are counted only once even if threatened by multiple fires. Unique feature identification uses:

  • Primary: Database unique identifiers (id, fid)
  • Fallback: Geometry coordinate hash for features without IDs

Visualization

Risk results are displayed through:

  • Map Overlays: High-risk features highlighted in red (fill opacity 0.6), moderate-risk in amber (fill opacity 0.4)
  • Statistical Summary: Sidebar panel displays counts of threatened features by category and risk level
  • Layer Control: Custom z-index panes ensure proper visual hierarchy (fires > analysis results > base layers)

Technical Stack

Frontend

  • Leaflet.js (interactive mapping)
  • Turf.js (spatial analysis)
  • Vanilla JavaScript (ES6 modules)

Backend & Database

  • Supabase PostgreSQL + PostGIS
  • Supabase Edge Functions (Deno runtime)
  • NASA FIRMS API integration

Data Processing

  • Python (GeoPandas, Pandas)
  • SQLAlchemy + GeoAlchemy2

Limitations and Assumptions

Temporal Considerations

  • Fire data represents 3-5 hour latency from satellite overpass to API availability
  • VIIRS sensor has twice-daily coverage; active fires may be missed between overpasses
  • Historical fire data not retained; analysis reflects current snapshot only

Spatial Accuracy

  • VIIRS nominal spatial resolution: 375m at nadir
  • Fire locations have inherent geolocation uncertainty (±200-500m typical)
  • Distance calculations use Euclidean distance; actual ground travel distance may differ due to terrain

Risk Assessment Caveats

  • Proximity-based risk does not account for: wind direction, fuel type, topography, fire behavior, or suppression efforts
  • Default distance thresholds (1000m high risk, 2500m moderate risk) are illustrative; users should adjust based on local conditions
  • Protected area analysis uses centroid approximation for large polygons, which may underestimate edge exposure

Citation

If you use this tool or methodology in your research, please cite:

Medina, P. (2026). Paraguay Wildfire Risk Explorer: A Real-Time Spatial Decision Support System for the Paraguayan Chaco. University of Maryland. Available at: https://chaco-wildfire-dashboard.paulo-medina.com/

Data Sources:

Contact

For questions, bug reports, or collaboration inquiries:
Paulo Medina
Email: pcmedina.avalos@gmail.com
GitHub: https://github.com/p-med

Paulo Medina - GEOG677